This is our project on Landmark Retrieval using 2 different models and to show how siamese network(similarity score output) outperforms a ResNet101 classification network for Landmark Retrieval.
The combination of the following GPUs were used for training from HPC Discovery Cluster Platform:
- NVIDIA Tesla A100
- NVIDIA Tesla P100
We haven't uploaded the trained model weights due to github size restrictions Siamese Model Weights: https://drive.google.com/file/d/1qeikdVI6KJ3Zj6FzQqcI3qSvi6kk8_HH/view ResNet-101 Baseline Model Weights: https://drive.google.com/file/d/1T0bCxtr0kV1GnYU8qZivzhQWKgv7wv6x/view?usp=sharing
The following is the retrieval results using the Siamese Network
The following is from the baseline ResNet101_Arcface Network
